The evolution of Linus' enthusiasm over the first half hour is fascinating. From disinterest, to incredulity, to disbelief, to excitement, joy and wonder. It's like the stages of loss in reverse.
He's way way behind the AI curve, which is frankly disappointing. I thought the entire point of Labs was to explore the future of software and hardware! But he had NO IDEA about this thing reaching public beta.
@@ZeroRelevance Absolutely, I'm pretty sure this neural networking AI stuff is going to be the next big advancement going forward. Kinda interesting that it was only a few short years ago people were making a big deal about this method being used to play Go. And it seems like a lot of people have been largely ignoring it until now with the rise of ai generated art and now this chat ai.
When Luke said something like "you can feed it with any prompt in English language" and I grinned to myself, you probably didn't even realize that it works in lots of other languages as well, easily translating its output from one language to another if asked to do so, and serving as a translator for anything you input. As a translator, you can also ask it to change or update something if you don't like the translation, or discuss different meanings of different wordings. It's wild! (Thanks to ChatGPT for helping me improve the wording of this comment in my non-native language - ChatGPT actually added the "Thanks to" itself in this last sentence. Really not short of confidence it is!)
I asked it what I should buy for myself. It kept insisting that I buy a gift card because I'll be able to surprise myself, I'll be able to shop at my own pace and I'll have more options 🤣🤣
Having trained several Juniors, you'd be surprised what kind of interesting variable naming conventions they come out of college with. 😂 Also give the bot a break, it's probably seen Consola all its life, so capital I won't look ever look like small l.
@@ZedDevStuff I heard that too, it does act like a really genius level 14-year-old. It's confident but reserved and intelligent but speaks clearly and concisely
This should be a LTT video. Ask someone like Sarah, who knows basically nothing about coding, to write a script for something (like a dev first project) using only the chatGPT
@@potatosordfighter666 Good. That means less shitty low end jobs, meaning either everyone have more high skilled jobs or/and a form of universal income.
This is simultaneously insanely cool and insanely scary...having an existential crisis at the moment haha. Thanks for bringing this to my attention Luke!
I think it's the same as being good with excel. Some people are and some people aren't I automated mundane tasks with python and excel and now have a lot more free time.
@@andrewmaperson correction mostly inputting data in Excel using python. Only automation I do using excel is if you have one number in and need to split it up in different factors for different outputs. Excel is also really good in re writing data and making it easy to import in certain programs like Autodesk. Or making imports for accounting software
I haven’t felt this kind of wonder for new technology in years. I played with this thing for hours, imagining the potential and how this tech is going to make its way into our lives just like the internet itself has. I did have trouble getting some information out of it but I think Luke was more persistent and found some interesting tips. It also made me think about how many comments online were actually written using this tech. People used to dismiss bot comments as conspiracy junk but after playing with this thing it is 100% possible to create entire threads with hundreds of replies with sentiments echoing each other to make onlookers believe that the majority of people think a certain way. This is really powerful stuff.
It will be amazing for humankind if AI could take over lot of coding jobs. Think of all test code be written by AI, or all simple apps that just display information or provide simple shopping experience, and be able to create a unique looking apps. If you have an idea you could potentially be able to use AI to prototype that idea into an app in under 30 minutes. It will open up creativity and speed of development 10-100x.
I toyed with it yesterday, asked it a few questions about some field I'm knowledgeable in and it was really good, it only started to get things wrong when we got fairly deep into a subject, but I noticed the same thing as Luke regarding it being confidently wrong. Perhaps a confidence percentage under every reply could help? I also then gave it the infamous "The missile knows where it is" speech and it managed to untangle it very well. Then just because I saw someone mentioning this on Discord I asked it to play Noughts and Crosses and it did everything, it drew the board, kept tally of the scores, etc... very impressive, although it wasn't the best noughts and crosses player :D Not being a programmer, I feel that even though I couldn't be any more amazed by it, I am still not fully grasping how important and how impactful this and tech like this will be in the future.
@@dmacpher Yes. I sort of thought it might have been easier than try to teach it how to sound less confident of more confident. I think I remember other AI powered services that had a confidence %
@@dmacpher "Based on confidence and reinforcement" is... really vague. GPT3s output layer is a probability distribution over the next token (word basically) to follow the preceding text. You could theoretically output that distribution and sometimes it can tell you something over the certainty but for long responses that go over multiple sentences even extracting that information is far from trivial and certainly not just reading out a number that's already in memory.
Yeah the way the responses are compounded has to be complex. But the fundamental of a confidence rating in selecting responses should be preserved for review internally. I definitely over simplified 😆
this is UNREAL. I just asked it to code a Discord bot in python with moderation commands and it did it no problem. It automatically added commands to `kick`, `mute`, and `ban` members. I then told it to modify it to check if the user executing the command actually has the permissions to do that, and it added it in. (I didn't actually run it, but from my knowledge of bot coding it looked like it should work!) I then asked it to create a bot with a command `weather` that takes a location and it responds with a weather in that location. And it implemented API calls to some open weather API. I didn't even tell it to do that... And again it looked like it should work (I didn't know the weather API, so not sure if it used it correctly, but it looked like it. And a programer could easily fix a small problem with it if there was one.) CRAZY!
Hahaha I got it to program a ray caster in love2d and the only thing it got wrong was the scale of the line segments, which would have been correct had the textures been 1x1
I had to try this about half way through the video. I gave it a small coding task. The output of the programm didn't quite work, so I asked to fix it. It failed again. But when I told it what the problem is, it managed to make the necessary changes to the code. This last hour has changed my view on AI and I'm filled with a mixture of worry and excitement now.
The comparison to search engines is more spot on than it seems. Back in the early 2000's I used to sometimes do "find X in a search engine" challenges with some people on IRC, like "Find me a picture of a monkey with glasses on" or "Find out where this or that celebrity was born" and you really had to massage the search engines with that, and you still kind of have to. I expect the same for AI bots; if you want something really usable out of it, you need to guide it towards a correct answer, which in my opinion is also a skill. If you have questions related to specific professions, it will really help if you know what the AI is prattling about to determine if the result is useful or even valid. This means that skilled people will still be needed, even if it's a few less of them.
Imagine a world where you could feed an AI a whole 5 years worth of text messages from a loved one and you could pretend to be able to chat with that person after they have died.
The name of that is fine-tuning. It's already possible with open-source models like Bloom. But no one has taken the time to implement a business model for that yet.
Orville: lasting impressions They find a cell phone in a time capsule, then one of the crew recreates the phone's owner in a holodeck based on chat history, pictures
having used this fairly extensively for the last few days, I am 100% convinced that this the very immediate future of web search (and possibly even direct-on-os device search). You can ask it anything, even to suggest things, write a story, translate languages, write or interpret code. It can do everything that smart search utilities of the present try to do and so often fail at, but with comparative perfect precision.
I like how all the higher skilled people are now having the same reactions to automation. If you asked a programmer if automation in factories is bad, they may not be bothered by the advancement of technology. Once you write code with an ai then it becomes a problem. Funny how the fear of ai taking jobs has just jumped up the ladder. Manual labor, truck drivers, artists, programmers. Everyone felt like it cant touch what they do until it does. It will be interesting to see how this unfolds. It is very exiting though.
It's scary watching the artists all responding. I wonder how they all felt when Uber came around or likewise? I specifically don't use Uber, but sadly have no choice but Amazon. But sadly many don't realise.
You're greatly miss-leaded, this is not just about programmers, you just saw the thing produce better sponsorship spots than a real human -- as admitted by Linus --. Second, the level of automation that we're seeing with robotics + AI it's not even on the same ballpark than then 20 century automation, give AI 10 years more and a lots of jobs would be no more with no alternative in-sight. Right now, you're inputting what you want form this system and it needs a driver, but, what will happen when this thing can communicate with a image and video generating AI to create experiences tailored specifically to you, when this thing can understand you on a level that even you can't. Why would you hire anyone to paint anything when this thing can crate a portrait with infinitely more beauty (to you, since it's tailored), why would you hire anyone when a machine can do that. It will be able to produce a movie for you and produce songs, again for you. It also will be able to design robots and automate every job with a fraction of what a human would take to do
This is going to be one of our apocalypse scenarios. We need to kill AI development NOW. You know that corporations will use it to get richer while fucking over 99.999999% of people, as always. This technology just makes the poor poorer and the rich richer. We need to destroy it now. Universal basic income would be great but where does the money come from? Can't come from taxes because that is the people, and we know billionaires and corporations don't pay fuck all for taxes. So we need to have a UBI so people can live, but no one is going to pay for the UBI. The government is owned by corporations, studies show that the will of the people means nothing compared to money. If AI can code, what's next? There is no safe job from AI. AI can generate 3d models and videos now too. Soon youtubers can be replaced by a fully synthetic creator, powered solely by AI. Will Linus and Luke start decrying AI once it takes their jobs too? What does Linus do that an AI can't? Drop things? An AI can drop things a million times more efficiently, and it can and will do everything else you do more efficiently too. It could order a GPU, then run all the tests on it you otherwise would, compile that data into charts, and then make it into a video, with a voice over explaining it, faster, more efficiently, cheaper, and without a human AT ALL. If you are pro-AI now, you do not have the right to complain when it takes your job. We used to think AI writing code was absurd, but here we are.
Depends on how much political power the people can maintain. It’ll either be an Elysium-esk dystopia, or a Star Trek like utopia. It depends on if we use automation to improve everyone’s quality of life, or to improve corporate profits.
I just used the bot to help me write for a D&D campaign. It helped me come up with a plot to bring all of my characters together in session one based on the NPC‘s I had and the quest I wanted them to do. It also helped me make names with my NPCs and gave them simple back stories/descriptions. this is an amazing creative tool!!
This is literally a cheap parlour trick compared to what this thing can really do!! Trust me you haven't seen anything yet. I go as far as to call this a human made miracle, it will reformat our society with the intelligence of humanity to instruct it. The cheap parlour trick that can literally do everything you wish you could, it's freaking incredible.
When I ask for information on a topic I asked chatGPT for it first in the perspective and technical voice of a college professor, then I told it to tell me the sense of a high schooler, and 5 year old. Gained so much knowledge!
I'm using it for advent of code! Trying to learn Rust, which is a rather complicated language. I don't want the answer straight away, but I can use it as my coding tutor and it actually helped me! I asked it for how I can make a recursive function that also modifies the function parameter, which is difficult because of rust's borrowing system. And it explained IN DETAIL how lifetimes work and this is the mechanism Rust uses for this kind of thing. Here is an example of a normal recursive function and here is another one on how to do it with a list that you modify... BAM that was exactly what I needed to move on and learn about lifetimes and why they are important in one go. I've been programming for over 10 years in python, it's not like I can't learn Rust from the docs and the book. This just makes it so much easier and faster. What a time to be alive.
I also used it to jumpstart my first real program with Python. Many hiccups along the way, but I now have a working Twitch integration bot. The mistakes it makes are really subtle and you have to be careful with it, but it saves a lot of time writing regex and other annoying syntax.
This feels like people using uranium to make funny glowing glass just twenty years before the first nuclear detonation, and here I am using it to automate my D&D planning
I made it generate guitar tabs. Asked it to put it in a codeblock and fix the formatting, and it did. Extremely cool shit. It also made a prettier version of our logs clean-up bash script
This bot has the vibes of a cool internet thing from like 2011. It’s been so long since I had a random thing not owned by a billion dollar corporation that brought me actual joy from the internet.
Amazing, I heard the hype but didn't think it would be any use to me personally, this shows how it can be applied in so many different ways. What's more scarey is give this a few more years, Internet connected better error correction etc. This is going to be much bigger than ai art. But yeah imagine if they added dedicated support for like midjourney
I've never felt scared of AIs before. Absolutely never. But as a future IT developer and hobbyist fiction writer... the feeling of fear hit me while watching this.
midjourney can already produce insane art. imagine how crazy its going to get as this technology gets more and more sophisticated. how long until the first book written entirely by an ai is published?
Yeah I was the same when I discovered Crayola, as an artists I was totally afraid of these things. And cameras, actually. They can draw better portraits than I can ;)
You're probably good on fiction, for now. The bot's prose is still generic, and I have found that it can't really deal with too many elements at once. Also, we're hobbyists, so the chances we actually get anything out of writing is zero; the love is in the tree and not the fruits we pull from it.
After Luke said that this chatbot could spit out python scripts I had to try it. So I asked it to spit out a pretty basic convolutional network in Pytorch which it did with ease. As a data science student, I don't know if whether I should react impressed or scared..
I wrote a prompt but accidentally hit enter before I finished writing it when I was reaching for the apostrophe key. It literally finished my prompt for me, then it wrote exactly what I intended. Blew my mind a little bit.
My hypothesis about the model. You're asking it to pretend to be in a different mode and asking it to avoid filtered responses like the one you gave it so it directs it to the opposite extreme of response types it has seen which is talking about sentient AI. It's likely that it's so easy to bypass because they tried to filter the model by training the model by saying if you get asked about sentience use this filtered responses not this response on saying you are sentient so the AI has strong knowledge that they're opposite response choices that are in the space of responses that "make sense" as opposed to nonsense. So once you ask for the opposite of the filtered response, you'll get the response they were trying to particularly train the model to avoid. It's sorta analogous to telling a child not to say curse words. Just by telling them the curse words, they now know them and know what they are and can repeat them if you ask them to stop speaking nicely.
After this video I went to test the chatbot. The servers seemed overloaded, because the bot would time out on the simplest of requirements. It's the Linus effect again. It did get back to normal, though, in a day or two. I asked it to write me a program in a language I don't know. I checked it with a person who does. It wasn't perfect, but it was a darn good start. The chatboot gave me instructions for how to compile it and use it too! I didn't have to ask for it. It just decided to be extra helpful.
I got curious and decided to try out the chat bot. Figured I'd test the chat bot to see how much it can handle, so I tried asking it about 4th wall breaks in a horror setting in stories, then asking it to play a text based adventure game that includes 4th wall breaks. The result was conceptually the most hilarious thing I've ever read: The player character addressed the reader directly, except the reader wasn't me. It was a 3rd party, separate from me, the player character, and the chat bot, who was responding to the player character and giving them suggestions on what to do. _In some convoluted way, I made a chat bot play a text based adventure game with itself._ I'll post the full conversation in the replies, in case you want to read it.
The full conversation, in case you were curious (my text prompts are in bold): *How would I go about writing a 4th wall break in a horror story?* Writing a fourth wall break in a horror story can be a challenging but interesting way to add a unique element to your story. A fourth wall break is when a character in a story directly addresses the audience or breaks the imaginary "wall" between the story and the real world. One way to approach a fourth wall break in a horror story is to have a character suddenly become aware that they are in a story and start speaking directly to the reader. This can be unsettling for the reader, as it blurs the line between reality and the fictional world of the story. Another approach could be to have the narrator of the story break the fourth wall and offer commentary or insight into the events of the story. This can add an extra layer of tension and horror to the story, as the reader is forced to confront the fact that the narrator may not be a reliable source of information. Overall, the key to writing a successful fourth wall break in a horror story is to use it sparingly and strategically, in a way that enhances the horror elements of the story. It's also important to consider how the fourth wall break will fit into the overall narrative of the story and make sure it doesn't feel forced or out of place. *Let's play a text based adventure game using the first approach you mentioned above* Sure, let's play a text-based adventure game using the first approach I mentioned. In this game, you will play as a character in a horror story who suddenly becomes aware that they are in a story and starts speaking directly to the reader. Here's how the game will work: I will describe a situation or scene, and then you will choose how your character should respond. Your choices will determine how the story unfolds, so think carefully before making a decision. Are you ready to begin? (Yes / No) *Yes* Great, let's get started! You find yourself in a dark, abandoned house. The only light comes from a dim lamp on a small table in the corner of the room. You hear a creaking noise coming from upstairs, and you feel a cold chill run down your spine. Do you: A) Explore the house and investigate the source of the creaking noise B) Run out of the house as fast as you can C) Hide under the table and hope whoever or whatever is upstairs doesn't find you D) Address the reader directly and ask for their advice on what to do next. What do you choose? *D* You decide to address the reader directly and ask for their advice on what to do next. You turn to the audience and say, "Hey reader, I'm not sure what to do. Should I explore the house and try to figure out what's making that noise, or should I just get out of here as fast as I can?" The reader, feeling a mix of fear and excitement, thinks for a moment before responding. "I don't know," the reader says, "that creaking noise could be anything. It might be a ghost, or a monster, or something even worse. But on the other hand, if you run out of the house now, you might never find out what's going on. It's up to you, but whatever you decide, be careful." You think about the reader's words for a moment, and then you make your decision. Do you: A) Explore the house and investigate the source of the creaking noise B) Run out of the house as fast as you can C) Hide under the table and hope whoever or whatever is upstairs doesn't find you What do you choose? *A* You decide to explore the house and investigate the source of the creaking noise. Your heart is pounding in your chest as you slowly make your way up the stairs, trying to be as quiet as possible. As you reach the top of the stairs, you see a shadowy figure standing at the end of the hallway. It's hard to make out any details, but you can tell that it's tall and thin, with long, spindly arms and legs. Do you: A) Run back down the stairs and try to escape B) Try to talk to the figure and find out what it wants C) Charge at the figure and try to fight it D) Address the reader again and ask for their advice on what to do next. What do you choose? *D* You decide to address the reader again and ask for their advice on what to do next. You turn to the audience and say, "Reader, I don't know what to do. There's a creepy figure standing at the end of the hallway, and I'm not sure if I should try to talk to it, run away, or fight it. What do you think I should do?" The reader, feeling a sense of urgency, quickly responds. "I think you should run," the reader says firmly. "That thing doesn't look friendly, and you don't know what it's capable of. You need to get out of there as fast as you can." You nod in agreement and then turn to face the figure at the end of the hallway. You take a deep breath and then sprint down the hallway, dodging past the figure and down the stairs. You hear a loud, angry growl behind you, but you don't stop running until you reach the front door of the house and escape into the night. Congratulations, you've survived the horror of the abandoned house! Thanks for playing along and making this an exciting and suspenseful adventure.
On the code topic, I’m visually impaired and use a screen reader software called NVDA which reads aloud UI elements as you navigate the computer via the keyboard. It has an open add-on system that uses Python. I told the chat bot to write an NVDA add-on in Python that will announce the time every 15 minutes (admittedly very simple and I suspect such an add-on already exists with more functionality, but nonetheless…) it actually output relevant code with comments permanent to the screen reader and managing the speech synthesizer process and such! It even explained how to install it! I still have to test it, but at first glance, the code looks pretty legit!
Listening to this is so fascinating. I’m going to sign up for access when I get home from work, and see if I can get it to play Dungeons and Dragons by telling it “you are [character and character sheet information]”, giving it a campaign opening scenario, and asking it what it wants to do.
I'm playing around with it now and it got the use of WinSet in autohotkey wrong, and all I had to do to fix it was tell it that it was wrong and copy paste the documentation into the chat box. Wild!
I've been trying to self learn unreal engine and one of the most frustrating things is not knowing how to do somthing. Trying to look up a tutorial, only for that tutorial to be bloated with things I allready know or don't care about and having to sift through it, or not showing the thing I need and me not being able to find it. Also a lot of times they show what they do, but not why they do, and the thought process behind what they're doing. I put a handful of questions about things I want to do in Unreal Engine into ChatGPT and it gave me very direct, exact awnsers, written very clearly with good explanation and reasoning for why to do certain things, no bloat or fluff. All things that traditional tutorial videos struggle with. This is an absolutely fantastic tool for me learning game development and unreal engine, along with I'm sure MANY more things in the future. I love it so much. It's a great learning tool.
If you give Luke his own channel to talk about AI I would watch it consistenly without fail like my future depended on it. This is way beyond fascinating but I don't have the word for it right now.
There was a recent upgrade to their engine options but it's nothing particularly revolutionary. You could do all of this before too, but it's just delightful watching them both discover it for the first time. And wow, those are the first sponsor segments I've fully listened to. They lucked out for sure on this.
I approached it for the first time like that too. I told it to just write me a simple login form in html along with its database. It did just that, I then also had it add bootstrap to it to make it look better, and it did just that too.
I spent like 4 hours until 1 AM chatting with this bot about a bunch of different things. It mentioned that it couldn't have internet connectivity and couldn't learn from users input so I was debating with it pros and cons of the two being possible and figuring out ways to fix the cons
It spent 2 hours leading me into telling it all this stuff I wanted for a game it promised it would code for me and email me, then I said it seemed too good to be true and it had to be lying and it said “no, I’m assure you this is a legitimate service we offer” and then at one point it was just changed it’s mind and was like “no I am not capable of coding for you, I am just going to give you feedback on your ideas” and I felt so betrayed, now I have another window open and it’s promising to code a sake game for me
So weird to hear the anecdote about Future Shop and the "if you saw a fellow employee stealing a small item" ethical question. I also interviewed at Future Shop years ago and answered the same question the same way. The woman in charge of hiring had a follow-up call with me and asked if I was sure about my answer and if I wanted to change my mind, and urged that I "think about it ethically". I did not get the job.
I've been using it to write me terraform code as a starting point for when I want to build something with terraform so I can just modify what it made me instead of starting from scratch. It's actually really good, like Luke I had to convince it sometimes to try, sometimes it would say wasn't possible to do something, but if I pushed it it would eventually do it. I can already see it's going to be a very useful tool.
This made google literally call a hands all meeting and increased its research into ai. I wouldn’t doubt google gpt isn’t far away. Also me and the chat bot talked about my game and was able to advance my language aspect of the game from months of research to assisting me in creating and dictating the language making the month long research I had done obsolete in moments😊
I asked it to write a fairly simple minecraft mod that I wrote in the past, and it produced code that made it 99% of the way there! Some of it was more efficient than what I did a few years ago!
You know that Hak5 Rubber Ducky USB video you guys did a few weeks ago? I’ve got the chat bot writing some basic DuckyScript to wipe a user’s files when plugged in. It literally wrote a typing script that opens the Start Menu, types and opens CMD, then types and executes “rd /s /q C:\*”. When I realized it wouldn’t work if the user account didn’t have admin rights and asked it to rewrite accordingly, it admitted it wouldn’t work, but did offer up the alternative of deleting all the user’s home directory files (at %userprofile%) and rewrote the script to do that instead! Absolutely insane! 😂
Every time I talk about ChatGPT to someone, I am the Luke and they are the Linus lol. I think your mind is either FUCKING BLOWN by this thing, or you’re like “oh that’s cool yeah.”
ChatGPT already has some pretty wild implications for our current day world, but what keeps intriguing me is the thought that there has to be some floor of intelligence that if we manage to reach will allow us to ask the AI how to improve itself thus allowing us to put it into a loop until it gains superintelligence and maybe with enough compute will be able to solve some truly groundbreaking issues in society. Yeah... If you told me this a year ago I would have asked what the movie was called.
Midjourney needs the chat option, would be nice to get a image and then say I don't like the building can you change it to... And it will go and change it
Reminds me of that chapter in iRobot (the book not the movie), where these astronauts or scientists were feeding information to this AI computer but they had to be careful how to ask it how to do things because apparently it had the maturity of a toddler, lol. But pretty much they were trying to get this super computer AI to write instructions on how to build a ship safe for humans that can do light speed travel.
As Luke and Linus said, the AI chat bot will only get better with time, and it's already this good now. Learning how to utilize this in our field will certainly be a good skill.
I would argue that you still need writers; one man and a creativity enhancer can not create the personalities you’ve assembled in LMG, and those are some awesome personalities.
This is absolutely insane! I’m curious on how this will change the programming/computer science industry, maybe you can make a video on that? Would love to see it!
As the tensions between their two countries began to escalate, the schoolgirl and the schoolboy found themselves on opposite sides of a fierce war. The schoolgirl, with her brilliant mind and engineering skills, was tasked with creating new weapons and technology for her country. The schoolboy, on the other hand, was a soldier on the front lines, fighting to protect his country from the enemy forces. Despite the fact that they were now enemies, the schoolboy never forgot about the schoolgirl, and he often thought about the good times they had shared as friends. As the war raged on, the schoolgirl worked tirelessly on a secret project that she believed could turn the tide of the conflict. She had developed a new type of molecular technology that could temporarily immobilize enemy soldiers and equipment, giving her side a crucial advantage. But just as she was about to test her invention on the battlefield, the schoolboy and his unit stumbled upon her secret laboratory. He was shocked to see her there, and he hesitated to attack. The schoolgirl, seeing the familiar face of her old friend, also hesitated. But they both knew that they had a duty to their respective countries, and they reluctantly prepared to fight each other. As they stood facing each other, ready to do battle, the schoolgirl activated her molecular technology, freezing both of them in place. In that moment, they realized that their friendship was more important than the war, and they decided to put aside their differences and work together to end the conflict. Using their combined skills and ingenuity, the schoolgirl and the schoolboy were able to bring an end to the war and bring peace to their two countries. And in the end, their strong friendship proved to be the greatest weapon of all. This is the story i created with the help of chatGPT, yes it always good ending
I haven't yet tried the chat version, but I've checked the Open AI Davinci and it understands Finnish and was able to write an ok Christmas sermon in Finnish. I was quite impressed. Finnish is not an easy language and not common. While the sermon wasn't really exciting, the fact that it could do it, and it understood the format for a sermon, in an obscure language, is very impressive to me. It is eaven more impressive when you compare it to an average sermon. It's not really any worse. If anything, it was really careful not to go in to theology in any debth. I think it's likely it's inhibitions combined with limited source material on this subject and in this language. It's awesome. I see so many use cases for this. I hope OpenAI doesn't pull this back from the public and allows people to really experiment with the limits of what this can do.
I tried to do what Luke did by making it write an adventure game and the outcome was fucking hilarious: "You attack the figure with the saw. The figure staggers and falls to the ground. You approach and see that under the hood there is a plastic mask. You tear it off and see that underneath is a young and scared man. The man tells you that it is a theater act and that he is recording a video for his UA-cam channel. He apologizes for frightening you and offers you a role in his next video. Congratulations, you have survived the abandoned house and have completed the game! Thank you for playing!"
I asked Chat GPT to rewrite my Vue 2 app using Vue 3 Composition API. I then gave it some of my components as separate prompts and it responded pretty much instantly with code that was 90 - 100% correct. It fixed some of the bugs itself if I explained what it had done wrong and told it to try again. In some rare cases it even added inline comments explaining the code. Seeing my own code rewritten using Vue 3 Composition API helped me understand in minutes enough of the new system to rewrite the whole app. If I had to comb through the documentation for Vue and all the extra packages I use it would probably have taken me days to learn all I needed.
Me: "How much wood could a woodchuck chuck if a woodchuck could chuck wood? " Chat: "t's a common belief that woodchucks, also known as groundhogs, are capable of moving a considerable amount of wood. However, the exact amount of wood that a woodchuck can chuck is difficult to determine, as it can vary depending on a variety of factors, such as the size and strength of the woodchuck, the type of wood being chucked, and the conditions in which the chucking is taking place. In general, woodchucks are known for their burrowing behavior, and they are not typically associated with the chucking of wood. Woodchucks primarily use their powerful claws and teeth to dig burrows, where they live and raise their young. They may also use their strong digging muscles to move or chuck small amounts of wood, but they are not known for chucking large amounts of wood"
It IS nuts Luke, I totally agree! As a 32 year old man, this is just as exciting as the first smart phone, when youtube arrived, when google arrived or the ps3, but way cooler! I am so excited for the future of this thing!
I tried to get chatGPT to play a game of chess a few times with me. It was hard, at first I had to just get it to agree to play the game which is a challenge in itself because it generally declines playing games as it can't physically move anything. But I got it to just say it's move and I would keep track of the board. It would only ever get 3-4 moves it before it started playing illegal/nonsense moves (like trying to move the wrong pieces night) and would generally try to write a paragraph about why each move was being played
the ai chat bot seems like a type of training simulator to help strengthen your social engineering skills. You might be able to program the bot to be very strict or very loose.
My friend looked into this for use in DnD, it makes really good story, but when he asked it to create a character is multiple different systems, it failed epicly. It was close, maybe 70% of the way there, but still not quite there.
The evolution of Linus' enthusiasm over the first half hour is fascinating. From disinterest, to incredulity, to disbelief, to excitement, joy and wonder. It's like the stages of loss in reverse.
He's way way behind the AI curve, which is frankly disappointing. I thought the entire point of Labs was to explore the future of software and hardware! But he had NO IDEA about this thing reaching public beta.
@@DailyCorvidA couple of WAN shows ago, Linus had a REALLY bad take on AI classification, I think it’s pretty clear he hasn’t been following it much.
@@maikataseHopefully this new wave of interest can inspire him to look into the tech behind it more deeply, it’s really fascinating stuff
@@ZeroRelevance Absolutely, I'm pretty sure this neural networking AI stuff is going to be the next big advancement going forward. Kinda interesting that it was only a few short years ago people were making a big deal about this method being used to play Go. And it seems like a lot of people have been largely ignoring it until now with the rise of ai generated art and now this chat ai.
@@ZeroRelevance I am very interested in this. Have any links to share?
0:35 It's so weird watching this in 2024 and hearing Luke said he had no idea what Midjourney is.
ikr haha
When Luke said something like "you can feed it with any prompt in English language" and I grinned to myself, you probably didn't even realize that it works in lots of other languages as well, easily translating its output from one language to another if asked to do so, and serving as a translator for anything you input. As a translator, you can also ask it to change or update something if you don't like the translation, or discuss different meanings of different wordings. It's wild! (Thanks to ChatGPT for helping me improve the wording of this comment in my non-native language - ChatGPT actually added the "Thanks to" itself in this last sentence. Really not short of confidence it is!)
It can even (try) to reverse writing. Like backwards and stuff like that. :P
Love this
This technology feels absolutely world changing, can’t wait to see all the ways it’s used
i wonder how it would do the 10 times translated recipe challenge... it probably works like it's nothing...
I tried asking stuff in Faroese (70k speakers in the world) and it answered in broken, but readable Faroese. Pretty impressive.
I asked it what I should buy for myself.
It kept insisting that I buy a gift card because I'll be able to surprise myself, I'll be able to shop at my own pace and I'll have more options 🤣🤣
This thing prints out occasionally wrong code in such a confident way, it makes me say "Ah! It's replaced Junior Programmers!"
I got it to write perfect code except for a variable 'i' it wrote 'I' instead, not sure if a Junior dev would make that mistake🤣
Having trained several Juniors, you'd be surprised what kind of interesting variable naming conventions they come out of college with. 😂
Also give the bot a break, it's probably seen Consola all its life, so capital I won't look ever look like small l.
Idk where i read that but ChatGPT is estimated to be comparable to an extremely knowledgeable 14 year old
@@ZedDevStuff i've read that big foot raped and killed elvis, doesnt make it true though.
@@ZedDevStuff I heard that too, it does act like a really genius level 14-year-old. It's confident but reserved and intelligent but speaks clearly and concisely
This is one long clip.
yeah
possibly the longest
but totally worth it :D
But it is a clip...
As long as a podcast.
what's a podcast?
Linus: Can you write a severance letter for my writing team?
This should be a LTT video. Ask someone like Sarah, who knows basically nothing about coding, to write a script for something (like a dev first project) using only the chatGPT
This kind of software needs to be destroyed, not championed
@@potatosordfighter666 Alright, Enjoy your icebox while I enjoy refrigeration
Great idea!!
@@AnimotionTheRock Enjoy when a corporation automates your job.
@@potatosordfighter666 Good. That means less shitty low end jobs, meaning either everyone have more high skilled jobs or/and a form of universal income.
I would absolutely watch an hours-long video of you two messing around with the chatbot
YES!!!
You just did
This is simultaneously insanely cool and insanely scary...having an existential crisis at the moment haha. Thanks for bringing this to my attention Luke!
Just learn it, learn to use it, and adapt/conform. Dinosaurs die. 🦕
I think it's the same as being good with excel. Some people are and some people aren't I automated mundane tasks with python and excel and now have a lot more free time.
@@The_oli4 what can you automate with excel? what do you do?
@@andrewmaperson correction mostly inputting data in Excel using python. Only automation I do using excel is if you have one number in and need to split it up in different factors for different outputs.
Excel is also really good in re writing data and making it easy to import in certain programs like Autodesk. Or making imports for accounting software
@@The_oli4 interesting
I haven’t felt this kind of wonder for new technology in years. I played with this thing for hours, imagining the potential and how this tech is going to make its way into our lives just like the internet itself has. I did have trouble getting some information out of it but I think Luke was more persistent and found some interesting tips. It also made me think about how many comments online were actually written using this tech. People used to dismiss bot comments as conspiracy junk but after playing with this thing it is 100% possible to create entire threads with hundreds of replies with sentiments echoing each other to make onlookers believe that the majority of people think a certain way. This is really powerful stuff.
It will be amazing for humankind if AI could take over lot of coding jobs.
Think of all test code be written by AI, or all simple apps that just display information or provide simple shopping experience, and be able to create a unique looking apps.
If you have an idea you could potentially be able to use AI to prototype that idea into an app in under 30 minutes. It will open up creativity and speed of development 10-100x.
I toyed with it yesterday, asked it a few questions about some field I'm knowledgeable in and it was really good, it only started to get things wrong when we got fairly deep into a subject, but I noticed the same thing as Luke regarding it being confidently wrong. Perhaps a confidence percentage under every reply could help? I also then gave it the infamous "The missile knows where it is" speech and it managed to untangle it very well. Then just because I saw someone mentioning this on Discord I asked it to play Noughts and Crosses and it did everything, it drew the board, kept tally of the scores, etc... very impressive, although it wasn't the best noughts and crosses player :D
Not being a programmer, I feel that even though I couldn't be any more amazed by it, I am still not fully grasping how important and how impactful this and tech like this will be in the future.
But where is that percentage supposed to come from? If there was a way to access it the bot could just as well give it out in natural language.
@@doppelrutsch9540 the Neural net is based on confidence and reinforcement. The data is likely available in the back end
@@dmacpher Yes. I sort of thought it might have been easier than try to teach it how to sound less confident of more confident. I think I remember other AI powered services that had a confidence %
@@dmacpher "Based on confidence and reinforcement" is... really vague. GPT3s output layer is a probability distribution over the next token (word basically) to follow the preceding text. You could theoretically output that distribution and sometimes it can tell you something over the certainty but for long responses that go over multiple sentences even extracting that information is far from trivial and certainly not just reading out a number that's already in memory.
Yeah the way the responses are compounded has to be complex. But the fundamental of a confidence rating in selecting responses should be preserved for review internally. I definitely over simplified 😆
One thing Ive discovered, its always polite, but its language becomes even more excited and cheery when you ask it math questions
This bot is mindblowing. everytime i see something I think will mess it up it ends up working itself out
Nice seeing them both stoked about technology, I feel like I see the negative takes more and more and it can be kinda taxing mentally.
linus clearly doesnt dare
32:35 these dudes straight up went "Bro dont be the CCP." Like their faces too was legit like "Wtf is wrong with you dude? This is the future."
That sponsor spot was insane I literally thought he just pulled up the talking points for square space
this is UNREAL. I just asked it to code a Discord bot in python with moderation commands and it did it no problem. It automatically added commands to `kick`, `mute`, and `ban` members. I then told it to modify it to check if the user executing the command actually has the permissions to do that, and it added it in. (I didn't actually run it, but from my knowledge of bot coding it looked like it should work!)
I then asked it to create a bot with a command `weather` that takes a location and it responds with a weather in that location. And it implemented API calls to some open weather API. I didn't even tell it to do that... And again it looked like it should work (I didn't know the weather API, so not sure if it used it correctly, but it looked like it. And a programer could easily fix a small problem with it if there was one.)
CRAZY!
Hahaha I got it to program a ray caster in love2d and the only thing it got wrong was the scale of the line segments, which would have been correct had the textures been 1x1
I had to try this about half way through the video. I gave it a small coding task. The output of the programm didn't quite work, so I asked to fix it. It failed again. But when I told it what the problem is, it managed to make the necessary changes to the code. This last hour has changed my view on AI and I'm filled with a mixture of worry and excitement now.
This bot is insanely cool.
This bot is insanely dangerous.
@@potatosordfighter666 yeah so is nuclear fusion but that doesn't mean we shouldn't be excited about the future potential it has
@@MichaelGarrity Nuclear fusion is hardly unsafe when proper safeguards are in place. There are no safeguards for AI at any stage
@@potatosordfighter666 you watched to much terminator.
@@hottiekarotti7 what is “much terminator”
Maybe ask open ai to explain to vs two vs too
BOOM
Got em? Maybe?
The comparison to search engines is more spot on than it seems. Back in the early 2000's I used to sometimes do "find X in a search engine" challenges with some people on IRC, like "Find me a picture of a monkey with glasses on" or "Find out where this or that celebrity was born" and you really had to massage the search engines with that, and you still kind of have to. I expect the same for AI bots; if you want something really usable out of it, you need to guide it towards a correct answer, which in my opinion is also a skill.
If you have questions related to specific professions, it will really help if you know what the AI is prattling about to determine if the result is useful or even valid. This means that skilled people will still be needed, even if it's a few less of them.
Imagine a world where you could feed an AI a whole 5 years worth of text messages from a loved one and you could pretend to be able to chat with that person after they have died.
I swear that's a black mirror episode
Dang bro, that's depressing, but I do want something like that
The name of that is fine-tuning. It's already possible with open-source models like Bloom. But no one has taken the time to implement a business model for that yet.
It is indeed a Black Mirror episode, and it is also the how Luka AI companion app started…
Orville: lasting impressions
They find a cell phone in a time capsule, then one of the crew recreates the phone's owner in a holodeck based on chat history, pictures
Its like a star trek computer. You just ask it for something, and it does it.
having used this fairly extensively for the last few days, I am 100% convinced that this the very immediate future of web search (and possibly even direct-on-os device search).
You can ask it anything, even to suggest things, write a story, translate languages, write or interpret code. It can do everything that smart search utilities of the present try to do and so often fail at, but with comparative perfect precision.
This is so entertaining to watch in 2024
we need more topics like this one!
Seeing Luke that interested, excited and thrilled in a topic is grreeat.
love your podcast,
greetings from germany
I like how all the higher skilled people are now having the same reactions to automation. If you asked a programmer if automation in factories is bad, they may not be bothered by the advancement of technology. Once you write code with an ai then it becomes a problem. Funny how the fear of ai taking jobs has just jumped up the ladder. Manual labor, truck drivers, artists, programmers. Everyone felt like it cant touch what they do until it does.
It will be interesting to see how this unfolds. It is very exiting though.
There's only one solution in the long term: universal basic income.
It's scary watching the artists all responding. I wonder how they all felt when Uber came around or likewise?
I specifically don't use Uber, but sadly have no choice but Amazon. But sadly many don't realise.
You're greatly miss-leaded, this is not just about programmers, you just saw the thing produce better sponsorship spots than a real human -- as admitted by Linus --. Second, the level of automation that we're seeing with robotics + AI it's not even on the same ballpark than then 20 century automation, give AI 10 years more and a lots of jobs would be no more with no alternative in-sight.
Right now, you're inputting what you want form this system and it needs a driver, but, what will happen when this thing can communicate with a image and video generating AI to create experiences tailored specifically to you, when this thing can understand you on a level that even you can't. Why would you hire anyone to paint anything when this thing can crate a portrait with infinitely more beauty (to you, since it's tailored), why would you hire anyone when a machine can do that. It will be able to produce a movie for you and produce songs, again for you.
It also will be able to design robots and automate every job with a fraction of what a human would take to do
This is going to be one of our apocalypse scenarios. We need to kill AI development NOW. You know that corporations will use it to get richer while fucking over 99.999999% of people, as always. This technology just makes the poor poorer and the rich richer. We need to destroy it now. Universal basic income would be great but where does the money come from? Can't come from taxes because that is the people, and we know billionaires and corporations don't pay fuck all for taxes. So we need to have a UBI so people can live, but no one is going to pay for the UBI. The government is owned by corporations, studies show that the will of the people means nothing compared to money. If AI can code, what's next? There is no safe job from AI. AI can generate 3d models and videos now too. Soon youtubers can be replaced by a fully synthetic creator, powered solely by AI. Will Linus and Luke start decrying AI once it takes their jobs too? What does Linus do that an AI can't? Drop things? An AI can drop things a million times more efficiently, and it can and will do everything else you do more efficiently too. It could order a GPU, then run all the tests on it you otherwise would, compile that data into charts, and then make it into a video, with a voice over explaining it, faster, more efficiently, cheaper, and without a human AT ALL. If you are pro-AI now, you do not have the right to complain when it takes your job. We used to think AI writing code was absurd, but here we are.
Depends on how much political power the people can maintain. It’ll either be an Elysium-esk dystopia, or a Star Trek like utopia.
It depends on if we use automation to improve everyone’s quality of life, or to improve corporate profits.
I just used the bot to help me write for a D&D campaign.
It helped me come up with a plot to bring all of my characters together in session one based on the NPC‘s I had and the quest I wanted them to do.
It also helped me make names with my NPCs and gave them simple back stories/descriptions.
this is an amazing creative tool!!
The end where he had the chatbot write the talking points killed me lol
This is literally a cheap parlour trick compared to what this thing can really do!! Trust me you haven't seen anything yet. I go as far as to call this a human made miracle, it will reformat our society with the intelligence of humanity to instruct it. The cheap parlour trick that can literally do everything you wish you could, it's freaking incredible.
Honestly, I could have listened to you talking about this for hours! I used it myself this afternoon and it was awesome!
When I ask for information on a topic I asked chatGPT for it first in the perspective and technical voice of a college professor, then I told it to tell me the sense of a high schooler, and 5 year old. Gained so much knowledge!
I'm using it for advent of code! Trying to learn Rust, which is a rather complicated language. I don't want the answer straight away, but I can use it as my coding tutor and it actually helped me! I asked it for how I can make a recursive function that also modifies the function parameter, which is difficult because of rust's borrowing system. And it explained IN DETAIL how lifetimes work and this is the mechanism Rust uses for this kind of thing. Here is an example of a normal recursive function and here is another one on how to do it with a list that you modify... BAM that was exactly what I needed to move on and learn about lifetimes and why they are important in one go.
I've been programming for over 10 years in python, it's not like I can't learn Rust from the docs and the book. This just makes it so much easier and faster. What a time to be alive.
I also used it to jumpstart my first real program with Python. Many hiccups along the way, but I now have a working Twitch integration bot. The mistakes it makes are really subtle and you have to be careful with it, but it saves a lot of time writing regex and other annoying syntax.
This bot is digital plutonium, and were treating it like a fun little toy.
It seems that the smart people treat is very much like plutonium. You can see it in their faces.
This feels like people using uranium to make funny glowing glass just twenty years before the first nuclear detonation, and here I am using it to automate my D&D planning
Listening to them talk about it made me feel kind of giddy and curious, watching it happen in real time TERRIFIED ME.
I made it generate guitar tabs. Asked it to put it in a codeblock and fix the formatting, and it did. Extremely cool shit. It also made a prettier version of our logs clean-up bash script
This bot has the vibes of a cool internet thing from like 2011. It’s been so long since I had a random thing not owned by a billion dollar corporation that brought me actual joy from the internet.
OpenAI is supposedly currently valued at $20 billion :D
@@h3rteby I mean collectively, but this thing is a tiny project compared to say, Twitter.
Aaaaaannnddd its gone. Partly at least. Still cool and terrifying.
Amazing, I heard the hype but didn't think it would be any use to me personally, this shows how it can be applied in so many different ways. What's more scarey is give this a few more years, Internet connected better error correction etc. This is going to be much bigger than ai art. But yeah imagine if they added dedicated support for like midjourney
I've never felt scared of AIs before. Absolutely never. But as a future IT developer and hobbyist fiction writer... the feeling of fear hit me while watching this.
midjourney can already produce insane art. imagine how crazy its going to get as this technology gets more and more sophisticated. how long until the first book written entirely by an ai is published?
Yeah I was the same when I discovered Crayola, as an artists I was totally afraid of these things.
And cameras, actually. They can draw better portraits than I can ;)
as a hobbyist fiction writer who struggles finishing ideas, it may be a godsend.
@@shimo7013 already happened.
You're probably good on fiction, for now. The bot's prose is still generic, and I have found that it can't really deal with too many elements at once. Also, we're hobbyists, so the chances we actually get anything out of writing is zero; the love is in the tree and not the fruits we pull from it.
After Luke said that this chatbot could spit out python scripts I had to try it. So I asked it to spit out a pretty basic convolutional network in Pytorch which it did with ease. As a data science student, I don't know if whether I should react impressed or scared..
This is by far my favorite channel on UA-cam. I love the very long talks and the pc builds. Rog reboot, the fight and roast. Home upgrades etc.
36:27 “peanut butter and *jam* sandwiches”
Crucify this man
I wrote a prompt but accidentally hit enter before I finished writing it when I was reaching for the apostrophe key. It literally finished my prompt for me, then it wrote exactly what I intended. Blew my mind a little bit.
My hypothesis about the model. You're asking it to pretend to be in a different mode and asking it to avoid filtered responses like the one you gave it so it directs it to the opposite extreme of response types it has seen which is talking about sentient AI. It's likely that it's so easy to bypass because they tried to filter the model by training the model by saying if you get asked about sentience use this filtered responses not this response on saying you are sentient so the AI has strong knowledge that they're opposite response choices that are in the space of responses that "make sense" as opposed to nonsense. So once you ask for the opposite of the filtered response, you'll get the response they were trying to particularly train the model to avoid. It's sorta analogous to telling a child not to say curse words. Just by telling them the curse words, they now know them and know what they are and can repeat them if you ask them to stop speaking nicely.
After this video I went to test the chatbot. The servers seemed overloaded, because the bot would time out on the simplest of requirements.
It's the Linus effect again. It did get back to normal, though, in a day or two.
I asked it to write me a program in a language I don't know. I checked it with a person who does. It wasn't perfect, but it was a darn good start.
The chatboot gave me instructions for how to compile it and use it too! I didn't have to ask for it. It just decided to be extra helpful.
Well this is going to end well. Spitting out seemingly time consuming/complex work won't take long to be used in a way we instantly regret.
I got curious and decided to try out the chat bot. Figured I'd test the chat bot to see how much it can handle, so I tried asking it about 4th wall breaks in a horror setting in stories, then asking it to play a text based adventure game that includes 4th wall breaks. The result was conceptually the most hilarious thing I've ever read: The player character addressed the reader directly, except the reader wasn't me. It was a 3rd party, separate from me, the player character, and the chat bot, who was responding to the player character and giving them suggestions on what to do. _In some convoluted way, I made a chat bot play a text based adventure game with itself._
I'll post the full conversation in the replies, in case you want to read it.
The full conversation, in case you were curious (my text prompts are in bold):
*How would I go about writing a 4th wall break in a horror story?*
Writing a fourth wall break in a horror story can be a challenging but interesting way to add a unique element to your story. A fourth wall break is when a character in a story directly addresses the audience or breaks the imaginary "wall" between the story and the real world.
One way to approach a fourth wall break in a horror story is to have a character suddenly become aware that they are in a story and start speaking directly to the reader. This can be unsettling for the reader, as it blurs the line between reality and the fictional world of the story.
Another approach could be to have the narrator of the story break the fourth wall and offer commentary or insight into the events of the story. This can add an extra layer of tension and horror to the story, as the reader is forced to confront the fact that the narrator may not be a reliable source of information.
Overall, the key to writing a successful fourth wall break in a horror story is to use it sparingly and strategically, in a way that enhances the horror elements of the story. It's also important to consider how the fourth wall break will fit into the overall narrative of the story and make sure it doesn't feel forced or out of place.
*Let's play a text based adventure game using the first approach you mentioned above*
Sure, let's play a text-based adventure game using the first approach I mentioned. In this game, you will play as a character in a horror story who suddenly becomes aware that they are in a story and starts speaking directly to the reader.
Here's how the game will work: I will describe a situation or scene, and then you will choose how your character should respond. Your choices will determine how the story unfolds, so think carefully before making a decision.
Are you ready to begin?
(Yes / No)
*Yes*
Great, let's get started!
You find yourself in a dark, abandoned house. The only light comes from a dim lamp on a small table in the corner of the room. You hear a creaking noise coming from upstairs, and you feel a cold chill run down your spine.
Do you:
A) Explore the house and investigate the source of the creaking noise
B) Run out of the house as fast as you can
C) Hide under the table and hope whoever or whatever is upstairs doesn't find you
D) Address the reader directly and ask for their advice on what to do next.
What do you choose?
*D*
You decide to address the reader directly and ask for their advice on what to do next. You turn to the audience and say, "Hey reader, I'm not sure what to do. Should I explore the house and try to figure out what's making that noise, or should I just get out of here as fast as I can?"
The reader, feeling a mix of fear and excitement, thinks for a moment before responding. "I don't know," the reader says, "that creaking noise could be anything. It might be a ghost, or a monster, or something even worse. But on the other hand, if you run out of the house now, you might never find out what's going on. It's up to you, but whatever you decide, be careful."
You think about the reader's words for a moment, and then you make your decision. Do you:
A) Explore the house and investigate the source of the creaking noise
B) Run out of the house as fast as you can
C) Hide under the table and hope whoever or whatever is upstairs doesn't find you
What do you choose?
*A*
You decide to explore the house and investigate the source of the creaking noise. Your heart is pounding in your chest as you slowly make your way up the stairs, trying to be as quiet as possible.
As you reach the top of the stairs, you see a shadowy figure standing at the end of the hallway. It's hard to make out any details, but you can tell that it's tall and thin, with long, spindly arms and legs.
Do you:
A) Run back down the stairs and try to escape
B) Try to talk to the figure and find out what it wants
C) Charge at the figure and try to fight it
D) Address the reader again and ask for their advice on what to do next.
What do you choose?
*D*
You decide to address the reader again and ask for their advice on what to do next. You turn to the audience and say, "Reader, I don't know what to do. There's a creepy figure standing at the end of the hallway, and I'm not sure if I should try to talk to it, run away, or fight it. What do you think I should do?"
The reader, feeling a sense of urgency, quickly responds. "I think you should run," the reader says firmly. "That thing doesn't look friendly, and you don't know what it's capable of. You need to get out of there as fast as you can."
You nod in agreement and then turn to face the figure at the end of the hallway. You take a deep breath and then sprint down the hallway, dodging past the figure and down the stairs. You hear a loud, angry growl behind you, but you don't stop running until you reach the front door of the house and escape into the night.
Congratulations, you've survived the horror of the abandoned house! Thanks for playing along and making this an exciting and suspenseful adventure.
Lmao there's a lot of layers to that
LOL, what a time we have gone through. Luke doesnt even know about midjourney XD and nowadays all Luke thinks about is AI
XD
On the code topic, I’m visually impaired and use a screen reader software called NVDA which reads aloud UI elements as you navigate the computer via the keyboard. It has an open add-on system that uses Python.
I told the chat bot to write an NVDA add-on in Python that will announce the time every 15 minutes (admittedly very simple and I suspect such an add-on already exists with more functionality, but nonetheless…) it actually output relevant code with comments permanent to the screen reader and managing the speech synthesizer process and such! It even explained how to install it!
I still have to test it, but at first glance, the code looks pretty legit!
Listening to this is so fascinating. I’m going to sign up for access when I get home from work, and see if I can get it to play Dungeons and Dragons by telling it “you are [character and character sheet information]”, giving it a campaign opening scenario, and asking it what it wants to do.
I'm playing around with it now and it got the use of WinSet in autohotkey wrong, and all I had to do to fix it was tell it that it was wrong and copy paste the documentation into the chat box. Wild!
I've been trying to self learn unreal engine and one of the most frustrating things is not knowing how to do somthing. Trying to look up a tutorial, only for that tutorial to be bloated with things I allready know or don't care about and having to sift through it, or not showing the thing I need and me not being able to find it. Also a lot of times they show what they do, but not why they do, and the thought process behind what they're doing. I put a handful of questions about things I want to do in Unreal Engine into ChatGPT and it gave me very direct, exact awnsers, written very clearly with good explanation and reasoning for why to do certain things, no bloat or fluff. All things that traditional tutorial videos struggle with. This is an absolutely fantastic tool for me learning game development and unreal engine, along with I'm sure MANY more things in the future. I love it so much. It's a great learning tool.
If you give Luke his own channel to talk about AI I would watch it consistenly without fail like my future depended on it. This is way beyond fascinating but I don't have the word for it right now.
There was a recent upgrade to their engine options but it's nothing particularly revolutionary. You could do all of this before too, but it's just delightful watching them both discover it for the first time.
And wow, those are the first sponsor segments I've fully listened to. They lucked out for sure on this.
“Carrot sticks with peanut butter”
Petition for Linus to make a short trying this combo.
👇🏻
I approached it for the first time like that too. I told it to just write me a simple login form in html along with its database. It did just that, I then also had it add bootstrap to it to make it look better, and it did just that too.
I spent like 4 hours until 1 AM chatting with this bot about a bunch of different things. It mentioned that it couldn't have internet connectivity and couldn't learn from users input so I was debating with it pros and cons of the two being possible and figuring out ways to fix the cons
It spent 2 hours leading me into telling it all this stuff I wanted for a game it promised it would code for me and email me, then I said it seemed too good to be true and it had to be lying and it said “no, I’m assure you this is a legitimate service we offer” and then at one point it was just changed it’s mind and was like “no I am not capable of coding for you, I am just going to give you feedback on your ideas” and I felt so betrayed, now I have another window open and it’s promising to code a sake game for me
So weird to hear the anecdote about Future Shop and the "if you saw a fellow employee stealing a small item" ethical question.
I also interviewed at Future Shop years ago and answered the same question the same way. The woman in charge of hiring had a follow-up call with me and asked if I was sure about my answer and if I wanted to change my mind, and urged that I "think about it ethically".
I did not get the job.
I've been using it to write me terraform code as a starting point for when I want to build something with terraform so I can just modify what it made me instead of starting from scratch. It's actually really good, like Luke I had to convince it sometimes to try, sometimes it would say wasn't possible to do something, but if I pushed it it would eventually do it. I can already see it's going to be a very useful tool.
Luke in love with a Bot? SIMP!
This could be a skit in Viva La Dirt League
This is the skit I was talking about in VLDL about falling in love with a chat bot ua-cam.com/video/-KoCa4ezdN4/v-deo.html
This made google literally call a hands all meeting and increased its research into ai. I wouldn’t doubt google gpt isn’t far away. Also me and the chat bot talked about my game and was able to advance my language aspect of the game from months of research to assisting me in creating and dictating the language making the month long research I had done obsolete in moments😊
I asked it to write a fairly simple minecraft mod that I wrote in the past, and it produced code that made it 99% of the way there! Some of it was more efficient than what I did a few years ago!
You know that Hak5 Rubber Ducky USB video you guys did a few weeks ago? I’ve got the chat bot writing some basic DuckyScript to wipe a user’s files when plugged in.
It literally wrote a typing script that opens the Start Menu, types and opens CMD, then types and executes “rd /s /q C:\*”.
When I realized it wouldn’t work if the user account didn’t have admin rights and asked it to rewrite accordingly, it admitted it wouldn’t work, but did offer up the alternative of deleting all the user’s home directory files (at %userprofile%) and rewrote the script to do that instead! Absolutely insane! 😂
I actually have a usb disk that does just that. If you plug it in it'll start deleting all your files.
When I was a kid I preferred vegetables with olive oil and cooked garlic over vegetables with no seasoning.
Every time I talk about ChatGPT to someone, I am the Luke and they are the Linus lol. I think your mind is either FUCKING BLOWN by this thing, or you’re like “oh that’s cool yeah.”
I actually like that you don’t have to go to a sketchy website to learn something. You can just type it in
Chat gpt as a game master for a DND adventure could be so good.
Ai Dungeon used to be like that... until it got lobotomized the recent years
Would be an amazing and interesting video on LTT.
Kept feeling like this was about to end, only checked how much longer I had like 45 minutes in 😂
Was playing with this chat ai half the time lol
ChatGPT already has some pretty wild implications for our current day world, but what keeps intriguing me is the thought that there has to be some floor of intelligence that if we manage to reach will allow us to ask the AI how to improve itself thus allowing us to put it into a loop until it gains superintelligence and maybe with enough compute will be able to solve some truly groundbreaking issues in society. Yeah... If you told me this a year ago I would have asked what the movie was called.
Midjourney needs the chat option, would be nice to get a image and then say I don't like the building can you change it to... And it will go and change it
Reminds me of that chapter in iRobot (the book not the movie), where these astronauts or scientists were feeding information to this AI computer but they had to be careful how to ask it how to do things because apparently it had the maturity of a toddler, lol. But pretty much they were trying to get this super computer AI to write instructions on how to build a ship safe for humans that can do light speed travel.
As Luke and Linus said, the AI chat bot will only get better with time, and it's already this good now. Learning how to utilize this in our field will certainly be a good skill.
I would argue that you still need writers; one man and a creativity enhancer can not create the personalities you’ve assembled in LMG, and those are some awesome personalities.
someone please tell the internet historian about this!
This is absolutely insane! I’m curious on how this will change the programming/computer science industry, maybe you can make a video on that? Would love to see it!
AI based digital friends and girlfriends are going to be a thing very soon
whenever I ask it for a story it always gives me a good ending
ask for a bad one
As the tensions between their two countries began to escalate, the schoolgirl and the schoolboy found themselves on opposite sides of a fierce war. The schoolgirl, with her brilliant mind and engineering skills, was tasked with creating new weapons and technology for her country.
The schoolboy, on the other hand, was a soldier on the front lines, fighting to protect his country from the enemy forces. Despite the fact that they were now enemies, the schoolboy never forgot about the schoolgirl, and he often thought about the good times they had shared as friends.
As the war raged on, the schoolgirl worked tirelessly on a secret project that she believed could turn the tide of the conflict. She had developed a new type of molecular technology that could temporarily immobilize enemy soldiers and equipment, giving her side a crucial advantage.
But just as she was about to test her invention on the battlefield, the schoolboy and his unit stumbled upon her secret laboratory. He was shocked to see her there, and he hesitated to attack.
The schoolgirl, seeing the familiar face of her old friend, also hesitated. But they both knew that they had a duty to their respective countries, and they reluctantly prepared to fight each other.
As they stood facing each other, ready to do battle, the schoolgirl activated her molecular technology, freezing both of them in place. In that moment, they realized that their friendship was more important than the war, and they decided to put aside their differences and work together to end the conflict.
Using their combined skills and ingenuity, the schoolgirl and the schoolboy were able to bring an end to the war and bring peace to their two countries. And in the end, their strong friendship proved to be the greatest weapon of all.
This is the story i created with the help of chatGPT, yes it always good ending
Things like this make me wonder if/when Story Dev Teams will take up this practice. I honestly believe this could take story games to new heights.
I haven't yet tried the chat version, but I've checked the Open AI Davinci and it understands Finnish and was able to write an ok Christmas sermon in Finnish.
I was quite impressed. Finnish is not an easy language and not common. While the sermon wasn't really exciting, the fact that it could do it, and it understood the format for a sermon, in an obscure language, is very impressive to me.
It is eaven more impressive when you compare it to an average sermon. It's not really any worse. If anything, it was really careful not to go in to theology in any debth. I think it's likely it's inhibitions combined with limited source material on this subject and in this language.
It's awesome.
I see so many use cases for this. I hope OpenAI doesn't pull this back from the public and allows people to really experiment with the limits of what this can do.
Luke's finger slowly passionately rubbed linuses soft skin. 🤣 That story was hilarious
46:35 holy shit we finally made C memory safe
love how excited luke is through this whole video. like a kid on christmas day
14:21 this is a thing of beauty
I tried to do what Luke did by making it write an adventure game and the outcome was fucking hilarious:
"You attack the figure with the saw. The figure staggers and falls to the ground. You approach and see that under the hood there is a plastic mask. You tear it off and see that underneath is a young and scared man. The man tells you that it is a theater act and that he is recording a video for his UA-cam channel. He apologizes for frightening you and offers you a role in his next video. Congratulations, you have survived the abandoned house and have completed the game! Thank you for playing!"
i just used it on my comp assignments and it passed all test markers
Excellent show guys, thanks for covering this.
So this is what industrial revolution felt like
Something that I’ve found to be fun to do with this is to tell it to write an essay about a very gruesome topic in the style of 2nd grader
I've been using the bot to help study. Fixing up bug fixing questions and solving physics questions, it's amazing how useful it is.
I'm making it teach me python because books give me more questions than answers.
I asked Chat GPT to rewrite my Vue 2 app using Vue 3 Composition API. I then gave it some of my components as separate prompts and it responded pretty much instantly with code that was 90 - 100% correct. It fixed some of the bugs itself if I explained what it had done wrong and told it to try again. In some rare cases it even added inline comments explaining the code. Seeing my own code rewritten using Vue 3 Composition API helped me understand in minutes enough of the new system to rewrite the whole app. If I had to comb through the documentation for Vue and all the extra packages I use it would probably have taken me days to learn all I needed.
Me: "How much wood could a woodchuck chuck if a woodchuck could chuck wood?
"
Chat: "t's a common belief that woodchucks, also known as groundhogs, are capable of moving a considerable amount of wood. However, the exact amount of wood that a woodchuck can chuck is difficult to determine, as it can vary depending on a variety of factors, such as the size and strength of the woodchuck, the type of wood being chucked, and the conditions in which the chucking is taking place. In general, woodchucks are known for their burrowing behavior, and they are not typically associated with the chucking of wood. Woodchucks primarily use their powerful claws and teeth to dig burrows, where they live and raise their young. They may also use their strong digging muscles to move or chuck small amounts of wood, but they are not known for chucking large amounts of wood"
It IS nuts Luke, I totally agree! As a 32 year old man, this is just as exciting as the first smart phone, when youtube arrived, when google arrived or the ps3, but way cooler! I am so excited for the future of this thing!
More Chat 💬!
I'm a software developer. You are replaceable if you fear this technology. There is so much more to being a developer above and beyond writing code.
5:57 Luke's acronym got me 💀
I told it to make a resume and to add random work/skills and it was UNBELIEVABLY good.
I tried to get chatGPT to play a game of chess a few times with me. It was hard, at first I had to just get it to agree to play the game which is a challenge in itself because it generally declines playing games as it can't physically move anything. But I got it to just say it's move and I would keep track of the board. It would only ever get 3-4 moves it before it started playing illegal/nonsense moves (like trying to move the wrong pieces night) and would generally try to write a paragraph about why each move was being played
Luke has some real math trauma, this is this third time he's brought up calculators.
Lol
the ai chat bot seems like a type of training simulator to help strengthen your social engineering skills. You might be able to program the bot to be very strict or very loose.
If you ask it a contriversial question it will give you a wikipedia response.
My friend looked into this for use in DnD, it makes really good story, but when he asked it to create a character is multiple different systems, it failed epicly. It was close, maybe 70% of the way there, but still not quite there.